Tools that run on your own machine, with no account and nothing uploaded. This page is for people who want narration to stay private, work without a connection, or avoid a cloud dependency. It covers one paid app and four open models.
The short answerChecked July 2026
Choose Vocast if
You want on-device voice cloning that behaves like a finished app, and you are on a recent Mac.
Choose an open model if
You are comfortable at a terminal, or you need Windows or Linux, and you do not mind assembling the workflow yourself.
Where local loses
No shared cloud voice library, and quality depends on your own hardware. A hosted service is still simpler for a large team.
The lineup, compared
Scroll the table sideways for more columns
VocastThis is us
Piper
Kokoro
XTTS v2
F5-TTS
Runs fully on device
✓
✓
✓
✓
✓
Nothing uploaded
✓
✓
✓
✓
✓
Finished app, no terminal
✓
–
≈community UI
≈community UI
–
Clones your own voice
✓
–
–
✓
✓
Price
$49 one-time
Free
Free
Free
Free
Runs on Windows or Linuxlimitation
–
✓
✓
✓
✓
Runs without a GPU
✓
✓
✓
≈slow on CPU
–
Commercial use clear
✓
✓
✓
≈check licence
≈check licence
On device means the model runs locally with nothing uploaded. Details are indicative as of July 2026; verify licences before commercial use.
The alternatives in detail
01
Vocast
This is us
On-device voice cloning and narration in a signed Mac app, with nothing uploaded.
$49 one-time
Strengths
+Cloning and rendering happen on your Mac
+No account, no server, no upload
+One-time price, handles long scripts
Trade-offs
–macOS on Apple Silicon only
–No stock voice library
–A focused language range
Best for
Mac creators who want privacy without a terminal
Verdict
The simplest way to get on-device cloning if you are on a recent Mac.
02
Piper
A small, fast local TTS engine with pre-trained voices, from Raspberry Pi to desktop.
Free, MIT
Strengths
+Runs almost anywhere, including low-power devices
+Clear MIT licence
+Cross platform
Trade-offs
–No voice cloning
–Command line only
–Flatter prosody on long passages
Best for
Offline speech inside an app or device
Verdict
The most portable local engine, if a stock voice is enough.
03
Kokoro
A compact open model with a natural read that runs comfortably on CPU.
Free, Apache 2.0
Strengths
+Good quality for its size
+No GPU required
+Permissive licence
Trade-offs
–No cloning of your own voice
–Needs Python and a model download
–Small voice set
Best for
Private batch narration on any machine
Verdict
The best CPU-only read here, if you do not need your own voice.
04
XTTS v2
A local cloning model that copies a voice from a few seconds of reference audio.
Free, restricted licence
Strengths
+Clones from a short sample
+Multilingual
+Runs offline once set up
Trade-offs
–GPU strongly recommended
–Restricted licence for commercial use
–Can drift on long passages
Best for
Local cloning experiments
Verdict
Strong local cloning, held back by the licence and the hardware it wants.
05
F5-TTS
A recent research cloning model with high quality and rough tooling.
Free, research licence
Strengths
+Cloning quality close to hosted services
+Fully local
+No cost at any volume
Trade-offs
–GPU required
–Research grade tooling
–No stable app
Best for
Technical users who want top local quality
Verdict
The highest local quality on this list, for people who enjoy the setup.